Undergraduate ThesisUndergraduate Thesis (Final Year Defense Project)January 2026
Unsupervised Anomaly Detection for Mass-Scale ECG Screening Using Spectral-Temporal Variational Autoencoders
Sourav Garodia, Minhazul Abedin
Developed an unsupervised deep learning system for large-scale ECG anomaly detection using advanced Variational Autoencoder architectures, achieving strong performance with a Spectral-Temporal VAE (F1-score: 0.90, AUC-ROC: 0.93). The work integrates multi-domain signal modeling with a complete end-to-end pipeline and a functional screening interface.
